The present invention relates to methods and apparatus for processing video signals for recording, including methods and apparatus for encoding digital video signals by means of discrete cosine transformation (DCT) for recording by a digital VTR.
Digital VTR's serve both to digitize a video signal and record the digitized signal on magnetic tape. The bandwidth of a digital video signal as sampled exceeds the practical recording capacity of the magnetic tape. Accordingly, it is impractical to record the digital video signal as sampled, so that the signal is first encoded by a highly efficient encoding process prior to recording.
It has been proposed to employ discrete cosine transformation in carrying out such a highly efficient encoding process for digital video signals to be recorded by a digital VTR. In the discrete cosine transformation process, the digital video data are first arranged in predetermined blocks. For example, such blocks can be composed of eight-by-eight picture elements or pixels in the time domain. The predetermined digital video blocks are transformed into frequency domain data by means of the discrete cosine transformation process.
The video signals possess correlation, so that upon transformation into the frequency domain, the resulting DC components are predominant. Moreover, the frequency components produced by discrete cosine transformation typically have their greatest power levels at the lowest frequencies and, as the frequencies of the components increase, the power levels of the components generally decrease.
Once the discrete cosine transformation process has been carried out, the frequency domain data is then encoded in a variable length code format, such as Huffman codes or the like. This serves to decrease the number of bits required to represent the transformed data. Where the data is to be recorded on magnetic tape, an error correction coding process using Reed Solomon codes typically is also carried out.
Since the frequency domain data is encoded in a variable-length format, the amount of data representing each video screen will vary from screen to screen. If the data is recorded in such variable amounts, the data representing each screen will not be coextensive with the recording tracks, so that editing of the recorded signals becomes difficult. In order to align the data of each screen with respective ones of the recording tracks, the quantization intervals employed in quantizing the variable-length encoded data are adjusted so that the amount of data representing each screen is maintained substantially constant.
Certain data patterns (such as a sky background, for example) change very little so that the values of the frequency coefficient data produced upon discrete cosine transformation of such patterns are quite small. Accordingly, when such data is variable-length encoded, a relatively small number of bits are produced and a relatively small quantization interval, therefore, will be selected for quantizing such data. However, in the case of data patterns having relatively large variations therein, the values of the frequency component data produced by discrete cosine transformation are relatively large. Consequently, a coarse quantization interval is selected for quantizing such data.
To prevent substantial variations in the quantization intervals among the various types of data, it has been proposed that a shuffling operation should be carried out so that the time sequence of the coefficient data for a given screen will not correlate with the spacial positions thereof. This conventional shuffling process employs a random selection of the data representing each screen.
However, the usefulness of data interpolation to counteract errors caused by head clogging or tape scratches was not foreseen when the conventional shuffling process was devised. Should one head of a digital VTR become clogged, the signals within a corresponding channel which would otherwise be reproduced thereby cannot be obtained. The presence of dirt or scratches on a tape guide can cause successive errors to occur in the longitudinal direction of a tape which passes over the guide. Since the data is shuffled randomly in the conventional process, the occurrence of such errors renders it the difficult to carry out interpolation to reconstruct the data which has been lost due to the error. A further disadvantage of random shuffling is the consequent difficulty in viewing reproduced screens in the course of a cue or review operation.